--- dataset_info: - config_name: arb_Arab features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 13564464469 num_examples: 2811912 - name: validation num_bytes: 3717211 num_examples: 2000 - name: test num_bytes: 149693703 num_examples: 29879 download_size: 11154765493 dataset_size: 13717875383 - config_name: ces_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 10802489660 num_examples: 2547927 - name: validation num_bytes: 2490065 num_examples: 2000 - name: test num_bytes: 132301431 num_examples: 29891 download_size: 9778697651 dataset_size: 10937281156 - config_name: cmn_Hani features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 13971653387 num_examples: 2868817 - name: validation num_bytes: 3500764 num_examples: 1997 - name: test num_bytes: 147842322 num_examples: 29910 download_size: 12321849079 dataset_size: 14122996473 - config_name: dan_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 9930320568 num_examples: 2321875 - name: validation num_bytes: 2598769 num_examples: 1996 - name: test num_bytes: 133823173 num_examples: 29866 download_size: 8830460093 dataset_size: 10066742510 - config_name: deu_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 12457654668 num_examples: 2816638 - name: validation num_bytes: 2824743 num_examples: 2000 - name: test num_bytes: 137849523 num_examples: 29917 download_size: 10997102606 dataset_size: 12598328934 - config_name: ell_Grek features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 10780509046 num_examples: 2225423 - name: validation num_bytes: 3846240 num_examples: 1997 - name: test num_bytes: 150872913 num_examples: 29743 download_size: 8804619850 dataset_size: 10935228199 - config_name: eng_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 12959824979 num_examples: 2637179 - name: validation num_bytes: 4244569 num_examples: 1997 - name: test num_bytes: 158739318 num_examples: 29841 download_size: 10875236278 dataset_size: 13122808866 - config_name: fas_Arab features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 9926038284 num_examples: 2105428 - name: validation num_bytes: 3365808 num_examples: 2000 - name: test num_bytes: 145128793 num_examples: 29894 download_size: 8161975675 dataset_size: 10074532885 - config_name: fra_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 12394472719 num_examples: 2804741 - name: validation num_bytes: 2853747 num_examples: 1999 - name: test num_bytes: 138109716 num_examples: 29908 download_size: 10882032614 dataset_size: 12535436182 - config_name: hun_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 9268430836 num_examples: 2164147 - name: validation num_bytes: 2629688 num_examples: 1997 - name: test num_bytes: 133453981 num_examples: 29833 download_size: 8298528019 dataset_size: 9404514505 - config_name: ind_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 12421264774 num_examples: 2742114 - name: validation num_bytes: 3188461 num_examples: 2000 - name: test num_bytes: 142543276 num_examples: 29838 download_size: 10674646850 dataset_size: 12566996511 - config_name: ita_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 11711452327 num_examples: 2747784 - name: validation num_bytes: 2691987 num_examples: 1999 - name: test num_bytes: 133549834 num_examples: 29734 download_size: 10419797688 dataset_size: 11847694148 - config_name: jpn_Jpan features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 13216643647 num_examples: 2887206 - name: validation num_bytes: 2989770 num_examples: 1995 - name: test num_bytes: 140318390 num_examples: 29952 download_size: 11477102872 dataset_size: 13359951807 - config_name: nld_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 11972681102 num_examples: 2805281 - name: validation num_bytes: 2500680 num_examples: 2000 - name: test num_bytes: 132840866 num_examples: 29885 download_size: 10648821733 dataset_size: 12108022648 - config_name: pol_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 12092305788 num_examples: 2819094 - name: validation num_bytes: 2561287 num_examples: 2000 - name: test num_bytes: 133696999 num_examples: 29903 download_size: 10880090264 dataset_size: 12228564074 - config_name: por_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 11793212859 num_examples: 2763318 - name: validation num_bytes: 2624846 num_examples: 2000 - name: test num_bytes: 134633636 num_examples: 29838 download_size: 10455879530 dataset_size: 11930471341 - config_name: rus_Cyrl features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 13929807991 num_examples: 2779122 - name: validation num_bytes: 4138051 num_examples: 2000 - name: test num_bytes: 155980669 num_examples: 29849 download_size: 11300461637 dataset_size: 14089926711 - config_name: spa_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 12028610659 num_examples: 2770292 - name: validation num_bytes: 2822097 num_examples: 1999 - name: test num_bytes: 137063924 num_examples: 29873 download_size: 10582910299 dataset_size: 12168496680 - config_name: swe_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 10845415215 num_examples: 2526425 - name: validation num_bytes: 2582123 num_examples: 1998 - name: test num_bytes: 133581160 num_examples: 29836 download_size: 9632035852 dataset_size: 10981578498 - config_name: tur_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 12219291931 num_examples: 2802936 - name: validation num_bytes: 2723209 num_examples: 1998 - name: test num_bytes: 135806515 num_examples: 29873 download_size: 10761051587 dataset_size: 12357821655 - config_name: vie_Latn features: - name: query dtype: string - name: text dtype: string - name: hard_negatives sequence: int64 - name: hard_negatives_score sequence: float64 - name: type dtype: string - name: query_token_count dtype: int64 - name: text_token_count dtype: int64 - name: id dtype: string - name: original_row_id dtype: int64 - name: date dtype: string - name: dump dtype: string - name: file_path dtype: string - name: score dtype: float64 splits: - name: train num_bytes: 10663033125 num_examples: 2210383 - name: validation num_bytes: 3703329 num_examples: 2000 - name: test num_bytes: 150559157 num_examples: 29882 download_size: 8868879812 dataset_size: 10817295611 configs: - config_name: arb_Arab data_files: - split: train path: arb_Arab/train-* - split: validation path: arb_Arab/validation-* - split: test path: arb_Arab/test-* - config_name: ces_Latn data_files: - split: train path: ces_Latn/train-* - split: validation path: ces_Latn/validation-* - split: test path: ces_Latn/test-* - config_name: cmn_Hani data_files: - split: train path: cmn_Hani/train-* - split: validation path: cmn_Hani/validation-* - split: test path: cmn_Hani/test-* - config_name: dan_Latn data_files: - split: train path: dan_Latn/train-* - split: validation path: dan_Latn/validation-* - split: test path: dan_Latn/test-* - config_name: deu_Latn data_files: - split: train path: deu_Latn/train-* - split: validation path: deu_Latn/validation-* - split: test path: deu_Latn/test-* - config_name: ell_Grek data_files: - split: train path: ell_Grek/train-* - split: validation path: ell_Grek/validation-* - split: test path: ell_Grek/test-* - config_name: eng_Latn data_files: - split: train path: eng_Latn/train-* - split: validation path: eng_Latn/validation-* - split: test path: eng_Latn/test-* - config_name: fas_Arab data_files: - split: train path: fas_Arab/train-* - split: validation path: fas_Arab/validation-* - split: test path: fas_Arab/test-* - config_name: fra_Latn data_files: - split: train path: fra_Latn/train-* - split: validation path: fra_Latn/validation-* - split: test path: fra_Latn/test-* - config_name: hun_Latn data_files: - split: train path: hun_Latn/train-* - split: validation path: hun_Latn/validation-* - split: test path: hun_Latn/test-* - config_name: ind_Latn data_files: - split: train path: ind_Latn/train-* - split: validation path: ind_Latn/validation-* - split: test path: ind_Latn/test-* - config_name: ita_Latn data_files: - split: train path: ita_Latn/train-* - split: validation path: ita_Latn/validation-* - split: test path: ita_Latn/test-* - config_name: jpn_Jpan data_files: - split: train path: jpn_Jpan/train-* - split: validation path: jpn_Jpan/validation-* - split: test path: jpn_Jpan/test-* - config_name: nld_Latn data_files: - split: train path: nld_Latn/train-* - split: validation path: nld_Latn/validation-* - split: test path: nld_Latn/test-* - config_name: pol_Latn data_files: - split: train path: pol_Latn/train-* - split: validation path: pol_Latn/validation-* - split: test path: pol_Latn/test-* - config_name: por_Latn data_files: - split: train path: por_Latn/train-* - split: validation path: por_Latn/validation-* - split: test path: por_Latn/test-* - config_name: rus_Cyrl data_files: - split: train path: rus_Cyrl/train-* - split: validation path: rus_Cyrl/validation-* - split: test path: rus_Cyrl/test-* - config_name: spa_Latn data_files: - split: train path: spa_Latn/train-* - split: validation path: spa_Latn/validation-* - split: test path: spa_Latn/test-* - config_name: swe_Latn data_files: - split: train path: swe_Latn/train-* - split: validation path: swe_Latn/validation-* - split: test path: swe_Latn/test-* - config_name: tur_Latn data_files: - split: train path: tur_Latn/train-* - split: validation path: tur_Latn/validation-* - split: test path: tur_Latn/test-* - config_name: vie_Latn data_files: - split: train path: vie_Latn/train-* - split: validation path: vie_Latn/validation-* - split: test path: vie_Latn/test-* --- # FineWeb2-IR Multilingual IR training set built by generating search-style queries from FineWeb-Edu (English) and FineWeb2-HQ (20 additional languages) using the model [hotchpotch/query-crafter-multilingual](https://huggingface.com/hotchpotch/query-crafter-multilingual). Each of the **21 languages** below has roughly **3M** (query, text) pairs plus validation (2k) and test (30k) splits. Use it for multilingual IR training, reranking, and evaluation. | Language | Subset ID | Train examples | | --- | --- | --- | | English | eng_Latn | 2,637,179 | | Russian | rus_Cyrl | 2,779,122 | | Chinese | cmn_Hani | 2,868,817 | | German | deu_Latn | 2,816,638 | | Spanish | spa_Latn | 2,770,292 | | Japanese | jpn_Jpan | 2,887,206 | | French | fra_Latn | 2,804,741 | | Italian | ita_Latn | 2,747,784 | | Portuguese | por_Latn | 2,763,318 | | Polish | pol_Latn | 2,819,094 | | Dutch | nld_Latn | 2,805,281 | | Indonesian | ind_Latn | 2,742,114 | | Turkish | tur_Latn | 2,802,936 | | Czech | ces_Latn | 2,547,927 | | Arabic | arb_Arab | 2,811,912 | | Persian | fas_Arab | 2,105,428 | | Hungarian | hun_Latn | 2,164,147 | | Swedish | swe_Latn | 2,526,425 | | Greek | ell_Grek | 2,225,423 | | Danish | dan_Latn | 2,321,875 | | Vietnamese | vie_Latn | 2,210,383 | ## Dataset construction 1. **Source corpora** - English: [HuggingFaceFW/fineweb-edu](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu). - Non-English: [HuggingFaceFW/fineweb-2](https://huggingface.co/datasets/HuggingFaceFW/fineweb-2) / FineWeb2-HQ. 2. **Cleaning & filtering**: remove rows with missing text, cap text length, keep passages whose token count (Qwen3-1.7B tokenizer) falls in a per-language range. 3. **Query generation**: - Prompt format: `system: language=\ntype=`, `user: `. - Model: `hotchpotch/query-crafter-multilingual` via vLLM (T=0, top-p=0.95, max_new_tokens=128). - Instruction types (7): `keywords`, `query`, `title`, `faq`, `summary`, `synonym_keywords`, `alt_query`. - Rotation weights: 5× for the first five types, 1× for `synonym_keywords` and `alt_query` (so these two are rarer by design). 4. **Post-filtering & split**: drop empty/"NONE" queries and queries ≥50 tokens (Greek cap 70); stratified split by `type`: validation 2,000, test 30,000, train = remainder. Columns kept: `query`, `text`, `type`, `query_token_count`, `text_token_count`, `id`, `original_row_id`, `date`, `dump`, `file_path`, `score`. 5. **Deduplication (no-dup release)**: within each split, remove any row whose `query` or `text` duplicates a previously seen `query` or `text` (after `strip()`), ensuring all queries and documents are unique within the split. ## Hard negatives (train/test) - For each language subset, all texts are embedded with BGE-M3 and indexed for ANN search. - Each query embedding retrieves the top 200 nearest texts **excluding the paired positive**; their IDs go to `hard_negatives` and cosine-similarity scores to `hard_negatives_score`. - The `score` column is the original FineWeb quality score, **not** a query–text similarity score. ## License This dataset follows the licenses and usage terms of its original sources: [HuggingFaceFW/fineweb-edu](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu) and [HuggingFaceFW/fineweb-2](https://huggingface.co/datasets/HuggingFaceFW/fineweb-2) / FineWeb2-HQ. These corpora are based on Common Crawl. Use and redistribution should follow the upstream dataset cards, Common Crawl terms, and any downstream requirements imposed by those corpora. The generated query fields are provided as part of this derived retrieval dataset and should be used together with the source-corpus terms.